Research On Improvement Of Gross Error Detection Method

In recent years, with the rapid development of integrated automation in process industry, the requirement of monitoring and optimization for process conditions has become more and more important. And because of the random errors and gross errors in the process, the data from instrumentscan’t meet the material and energy balances and is always deviated from the true value. These errors can directly affect the accuracy of parameter estimation, so the data rectification and gross error detection technique is applied to identify gross error, reconciles measured data, reduce the effect of random error and improve the quality of measured data.The paper mainly studies the gross error detection of the linear steady-state process, the research results are as follows:1) The theory and simulation of several classic gross error detection methods——the Global Test, the Measurement Test, and the Node Test is compared. The steps and characteristics of MT-NT and NT-MT methods are analyzed.2) The NT-MT gross error detection method based on F statistic is proposed. In this method, the merits of NT and MT methodsare applied,the traditional statistics is replaced with F statistic, and the covariance is rectified each time when a correspondent gross error was detected. The simulation results show that the small gross error can be easily detected in this way, and the efficiency is high,the result is good.3) The PC-yr-MT method is purposed based on eliminating the correlation between variables. The PC-yr method based on the principle component and the NT method are compared. The simulation results show that PC-yr can make up the defects of NT-MTwhen there are gross errors with similar magnitude.4) The new framework of module based on the least squares support vector regression is purposed from the point of changing the module. It can make up the defect of the traditional module based on the least square mathematical module and has a high detection level for linear steady system.